Systematic Knockdown of Epigenetic Enzymes Identifies a Novel Histone Demethylase PHF8 Overexpressed in Prostate Cancer with An

Home , JARID1B, KDM3A, KDM3B, PHF8

Oncogene (2012) 31, 3444–3456 & 2012 Macmillan Publishers Limited All rights reserved 0950-9232/12 www.nature.com/onc ORIGINAL ARTICLE Systematic knockdown of epigenetic enzymes identiﬁes a novel histone demethylase PHF8 overexpressed in prostate cancer with an impact on cell proliferation, migration and invasion

M Bjo¨rkman1,PO¨stling1,2,VHa¨rma¨1, J Virtanen1, J-P Mpindi1,2, J Rantala1,4, T Mirtti2, T Vesterinen2, M Lundin2, A Sankila2, A Rannikko3, E Kaivanto1, P Kohonen1, O Kallioniemi1,2,5 and M Nees1,5

1Medical Biotechnology, VTT Technical Research Centre of Finland, and Center for Biotechnology, University of Turku and A˚bo Akademi University, Turku, Finland; 2Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Helsinki, Finland and 3HUSLAB, Department of Urology, University of Helsinki, Helsinki, Finland

Our understanding of key epigenetic regulators involved in finger proteins like PHF8, are activated in subsets of specific biological processes and cancers is still incom- PrCa’s and promote cancer relevant phenotypes. plete, despite great progress in genome-wide studies of the Oncogene (2012) 31, 3444–3456; doi:10.1038/onc.2011.512; epigenome. Here, we carried out a systematic, genome- published online 28 November 2011 wide analysis of the functional significance of 615 epigenetic proteins in prostate cancer (PrCa) cells. We Keywords: prostate cancer; epigenetics; histone used the high-content cell-spot microarray technology and demethylase; migration; invasion; PHF8 siRNA silencing of PrCa cell lines for functional screening of cell proliferation, survival, androgen receptor (AR) expression, histone methylation and acetylation. Our study highlights subsets of epigenetic enzymes influencing different cancer cell phenotypes. Plant homeo domain Introduction (PHD) finger proteins have a key role in cell survival and histone methylation, whereas histone deacetylases were Cancer initiation and progression have been recognized primarily involved in regulating AR expression. In as complex processes that involve both genetic and contrast, JumonjiC-domain (JmjC) containing histone epigenetic alterations (Sharma et al., 2010). Epigenetic lysine demethylases (KDMs) mainly had an impact on cell changes in cancer progenitor cells may represent early or proliferation. Our results show that the KDMs JARID1B, initial steps in tumorigenesis, resulting in polyclonal PHF8, KDM3A, KDM3B and KDM4A were highly precursor populations prone to the accumulation of expressed in clinical PrCa samples. The PHD-finger additional genetic and epigenetic defects (Holst et al., protein 8 (PHF8), a transcriptional coactivator with both 2003). Furthermore, global changes in histone modifica- PHD- and JmjC-domains, was moderately to strongly tion patterns are functionally associated with cancer expressed in 80% of clinical PrCa samples, whereas 76% development and recurrence (Fraga et al., 2005; Seligson of normal and benign samples were negative or only et al., 2005, 2009). These epigenetic modifications are showed weak PHF8 expression. Strong PHF8 expression mediated by antagonizing sets of enzymatic complexes: correlated significantly with high Gleason grade and was the ‘writer’ proteins, which attach chromatin modifica- borderline significant for poor prognosis. The results of tions in a site-specific manner, and the ‘eraser’ proteins functional PHF8 knockdown implicate a role in cell that remove these (Ruthenburg et al., 2007). Such migration and invasion, as shown by cell motility and 3-D modifications are interpreted by ‘reader’ proteins that invasion assays. Our study suggests that various cellular specifically bind to the modified chromatin. These phenotypes are regulated by distinct subsets of epigenetic complexes direct transcriptional regulator complexes enzymes. Proteins interpreting and modifying histone to active promoter sites. Aberrant expression of methylation, such as JmjC-domain and particularly PHD- epigenetic targets, including DNA methyltransferases, methyl-CpG-binding proteins, histone lysine acetyl- transferases, histone deacetylases (HDACs), histone lysine methyltransferases, histone lysine demethylases Correspondence: Dr M Nees, Medical Biotechnology, VTT Technical (KDMs) and Polycomb group proteins mediate cancer Research Centre of Finland, and Center for Biotechnology, Institute for Molecular Medicine Finland and University of Turku and A˚bo development and progression (Ellis et al., 2009). A Akademi University, Turku, Varsinais Suomi FI-20520, Finland. spectrum of diverse DNA methyltransferase and HDAC E-mail: matthias.nees@vtt.fi inhibitors are currently in clinical development, whereas 4Present address: OHSU Knight Cancer Institute, Oregon Health & some have already been approved as anticancer drugs Science University, Portland, OR 97239, USA (Egger et al., 2004; Minucci and Pelicci, 2006), and 5These authors contributed equally to this work. Received 22 March 2011; revised 27 September 2011; accepted 2 October represent the first examples for a novel concept of 2011; published online 28 November 2011 therapeutics. PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3445 Prostate cancer (PrCa) represents the third most Ontology (GO) annotations related to epigenetics, chro- common cause of cancer deaths among Western men, matin remodelling/maintenance, and co-regulatory func- primarily because of the development of castration- tions. We included genes that contained at least one key resistant metastasizing cancer. Substantial evidence epigenetic protein domain (for example, SET, bromo, supports the involvement of epigenetic processes in JmjC, plant homeo domain (PHD), HDAC or chromo). PrCa progression (Seligson et al, 2005). At the DNA Additional candidates were selected based on literature level, multiple tumor suppressor genes involved in meta-searches and protein–protein-interaction data of regulation of prostate differentiation, proliferation or epigenetic transcriptional regulator complexes. This con- metastasis are effectively silenced by promoter methyla- certed approach defined a set of 615 human genes with tion; frequently detected already in premalignant lesions known or assumed epigenetic activity (Figure 1a, Supple- (Cooper and Foster, 2009). Furthermore, global chro- mentary Table 1). In combination with a selection of matin methylation and acetylation studies have revealed controls, each gene was targeted by two independent histone modification patterns that may be predictive for siRNAs, resulting in a library of 1328 siRNAs. This PrCa recurrence (Ellinger et al., 2010; Bianco-Miotto library was spotted on plastic cell culture carriers and et al., 2010). For example, increased global levels of screened by the cell-spot microarray technique (Rantala Histone H3 lysine 4 dimethylation and H3K18 acetyla- et al., 2010; Rantala et al, 2011) in VCaP cells. The impact tion are statistically linked to an increased risk of tumor of silencing on PrCa growth (Ki67), survival (cPARP) and recurrence (Bianco-Miotto et al., 2010). In line with this, expression of AR were addressed by high-content epigenetic enzymes interpreting and modifying histones immunofluorescence staining and imaging, using specific are often aberrantly expressed in PrCa (Cooper and antibodies (Figure 1b). Additionally, altered global levels Foster, 2009), such as class I HDACs, which are of H3K4 and H3K9 dimethylation and H3K18 and overexpressed in aggressive and hormone-refractory H4K16 acetylation, linked to poor cancer prognosis, were PrCa, thus contributing to androgen-independent, lethal studied (Fraga et al., 2005; Seligson et al., 2009) PrCa’s (Halkidou et al., 2004; Weichert et al., 2008). (Figure 1b). The functional effects of siRNAs was Eight KDMs are highly expressed in PrCa’s (Metzger considered significant if the difference of staining intensity et al., 2005; Kahl et al., 2006; Wissmann et al., 2007). exceeded a Z-score 4 þ 2 s.d., compared with the median Various HDACs and KDMs were shown to regulate of scrambled control siRNAs. This resulted in the androgen receptor (AR)-mediated transcription, thus identification of 272 siRNAs, targeting 231 genes affecting affecting gene expression patterns essential for PrCa the various endpoints (Supplementary Figure S1). GO proliferation, development and progression (Yamane enrichment analysis of primary hits was used to select et al., 2006; Wissmann et al., 2007; Wolf et al., 2007; significantly over-represented protein families for second- Xiang et al., 2007; Welsbie et al., 2009). Despite the ary screening (DAVID; http://david.abcc.ncifcrf.gov/). emerging role of epigenetic processes and large scale mapping of cancer-associated chromatin modifications in Clustering of cell-spot microarray screening data by PrCa, the functional significance of key ‘readers’, ‘writers’ portioning around medoids (PAM) and ‘erasers’ has not been systematically evaluated. In the Our large-scale primary screen was primarily designed present study, we have systematically screened the func- to identify epigenetic gene classes that significantly affect tional role of 615 epigenetic genes representing all known PrCa cell survival, proliferation, AR expression and and putative epigenetically active players in the human histone modifications, but at this stage not focused on genome by siRNA knockdown. Our multiplexed cell-based identification of single candidate genes. Therefore, the screens addressed proliferation, apoptosis, expression of Z-scores of hit siRNAs from all the seven immune AR, and global histone methylation and acetylation in staining markers were grouped into six clusters using PrCa cells. In these functional screens, we identified the portioning around medoids and hierarchical clustering family of JmjC-domain containing histone demethylases as (Figure 1c). To identify significantly enriched protein critical for PrCa proliferation and survival. These 32 domains, we functionally annotated each of the clusters histone demethylases and primarily JmjC-domain contain- separately against the InterPro protein domain database ing proteins were then further evaluated by a second, (http://www.ebi.ac.uk/interpro/) and EASE score focussed knockdown screen across a panel of normal and (Hosack et al., 2003), Supplementary Table 2, filtering malignant prostate and PrCa cell lines. We identified against the 615 input genes as background. This PHF8 to be highly expressed in PrCa and involved in cell approach revealed that some of the PHD-finger proliferation and motility, which was further validated by a domain-containing proteins were significantly enriched series of cell biological followup studies. in the cluster of siRNAs that reduce cell survival, and increase global H3K4 dimethylation (Figure 1c, Cluster I). As expected, the cluster of siRNAs reducing global Results H3K18 acetylation and H4K16ac levels contained most proteins with HDAC domains (Figure 1c, Cluster 2). HDACs and JmjC-domain KDMs regulate PrCa cell Interestingly, siRNA hits that strongly affected AR proliferation and AR protein levels expression and Ki67, were clustered in the same group. In order to cover the complete repertoire of epigenetic Another set of siRNAs targeting mainly JmjC-domain enzymes in the human genome, we have created an containing proteins significantly reduced cell proliferation epigenetic siRNA library, targeting genes with Gene and simultaneously upregulated global H3K9 dimethy-

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3446

Figure 1 Cell Spot Microarray experiments and siRNA library design. (a) Protein domain distribution of the genes included in custom epigenomic siRNA library. (b) The outline of the Cell Spot Microarrays (CSMA) technology. Each siRNA is spotted on the array with transfection reagent and cell attachment matrix by contact printing (upper left panel). Cells to be transfected are overlaid on the arrays and are only allowed to attach to the spots containing siRNA (blow up images of the spots, lower left panel). Immunoﬂuorescence staining and high content imaging at the siRNA spots allows multiplex functional analysis of marker proteins in response to epigenetic gene silencing (right panels). (c) Heat map showing Z-scores (o±2 s.d. from the global array mean) of the siRNA CSMA screen hits in VCaP cells. Six gene clusters (inset) were generated by partitioning around medoids (PAM) analysis; only clusters I–III are shown (full clusters in Supplementary Figure 1). Different end point measurements were grouped by hierarchical clustering and included Histone H3 lysine 9 dimethylation (H3K9me2), cleaved poly(ADP-ribose) polymerase (cPARP), Histone H3 lysine 4 dimethylation (H3K4me2), cell proliferation (Ki67), Histone H3 lysine 18 acetylation (H3K18ac), AR and Histone H4 lysine 16 acetylation (H4K16ac). Red denotes an increase in the Z-score reﬂecting the protein staining intensities whereas blue represents a reduction.

lation (Figure 1c, Cluster 3). A large group of siRNAs (Bjorkman et al, 2008) and JmjC-domain KDMs, as targeting high mobility group (Lange et al., 2008; Tang silencing of both protein classes significantly reduced et al., 2010) and a second set of PHD-finger proteins AR expression and cell proliferation. induced cell proliferation, upregulated AR, and increased global levels of H3K4 dimethylation and H3K18 Systematic analysis of all 32 histone demethylases- and and H4K16 acetylation (Cluster 6, Supplementary JmjC-domain-containing proteins identifies three histone Figure 1). Our primary screen thus indicated that demethylases important for PrCa cell proliferation the most promising classes of epigenetic enzyme Next, we selected all 32 genes containing histone for PrCa therapy may be represented by HDACs demethylase activity (for example LSD1) or contained

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3447 Table 1 Histone demethylases overexpressed in prostate tumours Gene Previous Ensmbl Gene ID Entrez N (normal N (cancer Mean Mean t-statistics P-value name ID samples) samples) normal cancer

JARID1B JARID1B ENSG00000117139 10765 147 329 426.1 518.2 5.56 5.69E-08 PHF8 PHF8 ENSG00000172943 731069 147 329 376.0 470.0 5.43 1.00E-07 KDM3A JMJD1A ENSG00000115548 NA 147 329 600.0 701.0 3.64 0.0003 KDM4B JMJD2B ENSG00000127663 23030 147 329 707.0 802.9 2.80 0.005 KDM4A JMJD2A ENSG00000066135 9682 147 329 312.0 354.9 2.28 0.02

The mRNA expression of 32 histone demethylases in 149 normal and 349 prostate adenocarcinoma samples was compared in our GeneSapiens human transcriptomics database and the significance of the overexpression was calculated with t-statistics. a JmjC-domain for a secondary siRNA screen, with the normal/benign samples were negative or showed only goal to further evaluate their functional significance in weak PHF8 expression, 80.3% of PrCa cases showed PrCa. A panel of malignant PrCa cell lines (MDA-PCa-2b, moderate to strong expression (Figures 3a and b). LNCaP, VCaP and DU-145) and one non-transformed Increased PHF8 expression significantly correlated with line (RWPE-1) were screened with a library (Supple- high Gleason grades (Figure 3c, Pearson w2 ¼ 0.03, mentary Table 3) containing four siRNAs to target each Spearman correlation þ 0.023) and was borderline gene, utilizing 384-well plate transfection protocols. significant in Kaplan–Meier survival analysis SiRNAs were considered a hit if they significantly (Figure 3d; P ¼ 0.06211). Interestingly, no adverse (Loess-scoreoÀ1.8 s.d. from the median) reduced events were observed in 273 PrCa’s with low/negative proliferation of PrCa, but not non-transformed PHF8 expression, whereas the PHF8-high patients had RWPE-1 cells in replicate screens, with a threshold of an overall 81% survival (240 months). PHF8 protein at least two siRNAs effective in at least one cancer cell was located in the nucleoli, in accordance with previous line (Supplementary Table 4). Concurrently, we ex- studies in vitro (Feng et al., 2010; Zhu et al., 2010). plored mRNA expression of all 32 histone demethylases Exemplary staining patterns for characteristic PHF8 in a set of 149 normal prostate and 349 prostate adeno- immunohistochemistry in normal prostate and low- or carcinoma samples, based on the GeneSapiens database high-Gleason grade PrCa samples are summarized in (Kilpinen et al., 2008, http://www.genesapiens.org/). Supplementary Figure 1. Higher magnification images, Five histone demethylases (KDM3A, KDM3B, indicating frequently observed heterogeneous staining KDM4A, JARID1B and PHF8) were identified as patterns and the nuclear localization predominant in significantly (Po0.05) overexpressed in PrCa versus metastases, are shown in Supplementary Figure 2. normal prostate. Of these, specific silencing of PHF8 also significantly reduced cell proliferation in the PHF8 regulates motility-related gene expression patterns secondary screens (Table 1). in PrCa To explore the functional consequences of high PHF8 PHF8 is overexpressed in PrCa and reduces PrCa cell expression in vivo cancer context, we studied PHF8 proliferation co-expressed genes in 233 prostate tumor samples using The genes JARID1B and KDM4A were described the GeneSapiens transcriptomics database (Kilpinen before as highly expressed in PrCa, and function as et al., 2008). In all, 757 genes correlated significantly coactivators of AR (Shin and Janknecht, 2007; Xiang (correlation X0.7, Po0.0001, FDR 0%) with PHF8 in et al., 2007). In contrast, PHF8 mRNA was not prostate tumors (Supplementary Table 5). Functional previously described as significantly (Po0.0001) over- GO annotation analyses (Hosack et al., 2003) revealed expressed in PrCa versus normal prostate. We therefore enrichment of genes linked to the central nervous system conducted a large-scale meta-analysis monitoring development and regulation of the actin cytoskeleton, altered PHF8 mRNA gene expression across selected, as well as axon guidance and regulation of the actin large-scale and publicly available gene expression cytoskeleton, both processes closely linked to cell studies, for example, the expression analysis in Oncol- motility (Supplementary Table 6). This was further ogy project (Figure 2a), the recent comprehensive validated by the identification of genes functionally clinical PrCa profiles from MSKCC (Taylor et al., linked to cellular motility processes like integrins 2010; Figure 2b), and an older study (Chandran et al., ITGB2, ITGAM, ITGA9, ITGAB2, fibronectin, RhoQ 2007, Figure 2c). Furthermore, we studied PHF8 and ROCK kinase 1. To experimentally verify this expression in our GeneSapiens database (Kilpinen finding, we studied altered gene expression in LNCaP et al., 2008) across 496 PrCa samples (Figure 2d). In cells as the result of PHF8 silencing. PHF8 silencing all analyzes, PHF8 was significantly overexpressed in significantly reduced the expression of PHF8 mRNA PrCa compared with normal/benign samples. Next, we after 48 h posttransfection, as shown by quantitative studied PHF8 protein expression using immune-histo- RT–PCR (Figure 4a) and western blot (Figure 4b). chemical staining in tissue microarrays (TMAs) contain- A small set of 62 genes was significantly downregulated ing a total of 1179 cores, corresponding to 332 cancer and 22 genes were upregulated by PHF8 knockdown and 90 normal/benign specimens. Although 76% of (fold change±1.4, Po0.05) (Volcano plot; Figure 4c,

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3448

Figure 2 Analysis of PHF8 mRNA gene expression in large-scale transcriptomics data sets. (a) Expression of PHF8 in the expO (expression project for Oncology) data set. (b) the Memorial Sloan Kettering Cancer Center (MSKCC) PrCa data set and (c) transcriptome analyses of normal prostate and malignant PrCa samples by Chandran et al., 2007. (d) Meta-analysis of PHF8 mRNA expression across 149 healthy and 349 tumor samples in the ‘GeneSapiens’ human transcriptomics database. The box limits depict lower and upper 90% quartiles of the expression data and the red and green lines the median of expression. A full colour version of this ﬁgure is available at the Oncogene journal online.

Supplementary Table 7). The observed ratio of few PHF8 also had significant, but not dramatic effects on upregulated to many downregulated genes was in line apoptosis in LNCaP cells (Figure 4d). with the suggested positive coactivator function of PHF8 (Fortschegger et al., 2010). Functional annota- PHF8 mediates cancer cell migration and invasion tion of the 62 downregulated genes showed significant As modulation of the actin cytoskeleton and integrin enrichment of the integrin signaling pathway. This signaling are linked to cellular movement (Guo and correlates with our in silico findings, which linked Giancotti, 2004), we hypothesized that PHF8 may have PHF8 to gene expression patterns related to cell motility a role in PrCa cell migration and invasion. Cell (Supplementary Table 6). Functional knockdown of migration and invasiveness was studied in PC-3 cells,

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3449

Figure 3 Immunohistolological staining of PHF8 expression in benign and PrCa samples by TMAs. (a) PHF8 protein is absent or weakly expressed in 76% of normal and benign prostate samples, but moderately to strongly in 80.3% of malignant PrCa biopsies (b). (c) Moderate to strong expression of PHF8 is correlated to high-grade Gleason score tumors. (d) Strong/moderate expression of PHF8 correlates with poor patient survival (81% cumulative survival, borderline signiﬁcance P ¼ 0.0621). None of the PrCa’s with predominant negative or weak PHF8 staining showed any adverse events (100% survival, 300 months).

Figure 4 Effects of siRNA silencing of PHF8 in PrCa cell lines. (a) Real-time RT–PCR validation of silencing effect for three independent siRNAs in LNCaP cells. (b) The same siRNAs also significantly affect the levels of PHF8 protein. Multiple transcripts are characteristic for PHF8 and represent splice variants. (c) Bioinformatic analysis of transcriptional effects of PHF8 silencing on genome-wide mRNA gene expression patterns (volcano plot to visualize the most significant target genes, details in the text). (d) Silencing of PHF8 roughly doubles the amount of apoptotic cells in the LNCaP line. *Po0.05; **Po0.01. A full colour version of this figure is available at the Oncogene journal online.

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3450

Figure 5 PHF8 regulates cell migration, motion paths and invasive properties in 2-D and 3-D organotypic culture. (a) Quantiﬁcation of migration as in A for mock and HA-PHF8 transfected PC-3 cells expressing a tagged but fully functional PHF8 protein. ( þ s.d., n ¼ 3, *Po0.001). (b) Quantiﬁcation of cell migration and motion paths by time-lapse microscopy of scrambled and siPHF8 transfected PC-3 cells for 16 h. 15–20 cells/treatment were followed ( þ s.d., n ¼ 3, *Po0.001). (c) Representative images for scramble and siPHF8 transfected PC-3 organotypic spheroids after 8 days of growth in laminin-rich extracellular matrix (Matrigel). Confocal images were analyzed by the automated image analysis program (ACCA) automated image analysis software (lower panel). Red ¼ apoptotic cells, green ¼ SYTO62 stained live cells. (d) Quantitative analysis of image data, monitoring the growth of the spheroids (area ¼ size; top panel), loss of the round spheroid phenotype (measured as percentage roundness, inversely correlated to the formation of invasive structures; middle panel) and the number/ratio of apoptotic cells inside the spheroids; bottom panel.

because these were the most motile/invasive cells in both Additionally, the role of PHF8 in cell invasion was 2-D monolayer and 3-D organotypic cultures (Harma studied using 3-D organotypic cell culture models et al., 2010). PC-3 cells expressed similar levels of PHF8 (Harma et al., 2010). This method allows monitoring as LNCaP, which are considerably less invasive in 3-D. the effect of gene silencing on morphology, polarization PHF8 was thus silenced by siRNA and in parallel and growth of PrCa spheroids, which typically corre- overexpressed, using an HA-tagged expression vector sponds to the remaining differentiation potential of that contains the PHF8 open reading frame under the cancer cells. 3-D assays also allow the simultaneous control of the CMV promoter. Cumulative cell move- measuring of both cell growth ( ¼ spheroid size) and ment and motion paths were tracked for a period of 16 h invasion ( ¼ formation of invadopodia, degradation of with a phase contrast microscope. PHF8 overexpression basement membrane, loss of rounded spheroid shape of PHF8 significantly increased (Po0.001) the median and overt migration of cells into the extracellular length of the migration path (Figure 5a). Concordantly, matrix). PHF8 expression was silenced by RNAi for silencing significantly decreased the median length of the 68 h; the cells were then transferred into Matrigel and cell migration path (Po0.001) (Figure 5b). These data imaged for 8 days. Phase contrast images (Supplemen- suggests that PHF8 supports the regulation of cell tary Figure 3) were analyzed with VTT’s proprietary migration and motility, possibly via its effects on automated image analysis program ACCA developed expression of genes like integrins, ROCK kinase and for dynamic morphometric measurements in multi- others (Supplementary Table 6). In contrast, in scratch cellular spheroids. PC-3 spheroids first form invasive wound/wound healing cell migration assays, PHF8 filopodia (invadopodia) 6 days after seeding (Figure 5c). knockdown showed only minor effects on cellular PHF8 silencing significantly (Po0.0001) reduced both motility (Supplementary Figure 4). the spheroid size (‘area’) and the number of invasive

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3451

Figure 6 PHF8 regulates prostate cell proliferation. (a) The effect of PHF8 silencing on cell proliferation of a panel of normal and malignant prostate cell lines was measured with CellTiter-Glo luminescent assay 72 h after siRNA transfection and compared with scramble siRNA transfected control (±s.d., n ¼ 3*Po0.001). (b) The effect of PHF8 silencing on cell cycle was studied using propidium iodine staining followed by FACS analysis in LNCaP cells 72 h after transfection (±s.d., n ¼ 3, *Po0.02). (c) Proliferation of LNCaP PrCa cells with or without PHF8 silencing was analyzed in real time (measurements 1 image/h) for a period of 7 days. cellular structures (‘appendages’), and the spheroids Discussion remained in a round shape (% roundness; Figure 5d). In contrast, spheroids transfected with the scrambled Our systematic high-content functional screen covered control siRNA (left) and the poorly effective siPHF8_2 most known epigenetic enzymes and regulators with lose the round shape and form invadopodia and invasive several functional phenotypic endpoints, and resulted in structures. Furthermore, the most effective siPHF8_1 the identification of distinct classes of epigenetic also significantly increases the number of apoptotic cells modifiers as critical for cancer biology. The use of inside the spheroids. These data demonstrate that PHF8 multiple endpoint measurements indicated overall pro- regulates proliferation and invasive phenotypes not only tein-domain specific effects, and pointed to a prominent in classical 2-D cell culture models, but also more role of, for example, PHD fingers in cell survival and prominently in 3-D organotypic PrCa cell models. H3K4 dimethylation. In contrast, HDACs were primarily functional in the regulation of the AR expression, PHF8 mediates prostate cell proliferation whereas JmjC-domain KDMs mainly regulated cell As PHF8 has been previously linked to rRNA synthesis proliferation. Overall, this analysis demonstrates the and assigned a putative function in cell cycle progression distinctive value of genome-wide, systematic functional (Feng et al., 2010; Liu et al., 2010; Zhu et al.,2010),weset knockdown screens, addressing all known epigenetic out to validate these effects in PrCa cell lines. PHF8 enzymes in parallel. These studies have the potential to silencing significantly (Po0.001) reduced cell proliferation identify the most potent regulators as well as highlight of the most rapidly proliferating PrCa cell lines (LNCaP, their functional importance not only as single genes but 22rV-1, DU-145 and PC-3), and least affected the non- also as distinct groups. Silencing of a specific set of transformed normal primary prostate epithelial and pre- PHD-finger genes in PrCa cells increased both global malignant RWPE-1 cells (Figure 6). We further analyzed levels of H3K4 dimethylation and apoptosis. Over 200 the effects of PHF8 silencing on cell cycle progression, PHD-finger proteins exist in the human genome, which using propidium-iodine staining and FACS analysis interact specifically with trimethylated histone H3K4 (Figure 6b), which showed only minor but statistically and K36, thus interpreting and mediating methylation significant changes in (Po0.01) of LNCaP cells in the G1 events into epigenetic programs (Mellor et al., 2006; Shi phase after PHF8 silencing. The most striking differences et al., 2007). Aberrant expression of several PHD-finger are revealed in long-term growth analyses using the proteins is associated with a wide range of human IncuCyte real-time imaging device (Essen Bioscience, Ann pathologies, including cancers and neurological diseases Arbor, MI, USA) (Figure 6c). (Baker et al., 2008). Histone H3K4 di- and trimethyla-

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3452 tion sites are enriched at actively transcribed promoter strong mediator of cancer cell migration and invasion and enhancer chromatin regions (Ernst and Kellis, (Figure 4), and has weak anti-apoptotic functions. A 2010), and regulate gene expression particularly in more detailed analysis of potential PHF8 target genes development (Ruthenburg et al., 2007). Our screen involved in PrCa cell migration may be an important suggests that PHD-finger proteins may also be involved area for future investigation, and to elucidate the in the observed global loss of Histone H3K4 dimethyla- functional mechanisms PHF8 is involved in. Interest- tion, associated with tumor recurrence in PrCa patients ingly, PHF8 was found to bind to the integrin beta-1 (Seligson et al., 2005). promoter, and linked to positive regulation of genes Our study also identified HDACs as the main related to focal adhesion in HeLa cells (Fortschegger epigenetic protein class resulting in reduced AR protein et al., 2010). PHF8 silencing also reduced proliferation levels upon silencing. This is in line with previous and mildly inhibited G1-S phase transition, which is in observations, indicating that HDAC inhibitors potently line with a suggested role of PHF8 in cell cycle reduce AR protein expression, and downregulate progression (Feng et al., 2010; Liu et al., 2010). androgen-responsive genes (Bjorkman et al., 2008; However, a functional role of PHF8 in cancer metastasis Welsbie et al., 2009). Overall, these data suggest that and invasion needs to be further verified in relevant HDACs have a specific role in regulating AR activity in vivo models. and AR signaling in castration-resistant PrCa. Thus, Taken together, our findings suggest again that HDAC inhibitors could become an important therapeu- effector proteins like PHF8, which are involved in tic option for advanced PrCa patients in the future. interpreting and modifying histone methylation marks, Furthermore, we identified the JmjC-domain contain- may have a significant role in PrCa pathogenesis and ing protein family as critical for PrCa cell proliferation. progression. Several epigenetic cancer therapies target- JmjC proteins were enriched among those siRNAs that ing histone acetylation and DNA methylation have been significantly reduced Ki67 staining. We further validated established in the clinics, and the enzymes or protein these results by a secondary screening, targeting 32 complexes involved in modifying histone methylation KDMs and JmjC-domain containing proteins in a panel are considered to be highly attractive drug targets. of PrCa cell lines and non-transformed normal epithelial KDMs can be non-specifically targeted by amine cells. This experimental approach was combined with an oxidases inhibitors, which are currently indicated as in silico analysis of JmjC protein expression across antidepressants (Lee et al., 2006). Recent reports showed several large-scale clinical transcriptome data sets, that PrCa cell lines respond to MAOA inhibitors, linking for the first time the histone demethylase possibly via inhibition of KDM’s (Flamand et al., PHF8 with progression of PrCa. PHF8 was previously 2010). Furthermore, the JmjC-domain containing identified to demethylate mono- and dimethylated KDMs including PHF8 were found to be repressed by histone H3K9 and monomethylated histone H3K20 compounds targeting the Fe (II)/a-ketoglutarate depen- (Horton et al., 2010; Loenarz et al., 2010; Qi et al., dent oxygenases (Cloos et al., 2006; Smith et al., 2007; 2010). Inactivating mutations in the PHF8 JmjC- Rose et al., 2008). Taken together, such compounds and domain are linked to mental retardation syndromes epigenetic therapeutic concepts may offer intriguing with the cleft lip/palate (Laumonnier et al., 2005; Abidi starting points for the development of novel epigenetic et al., 2007; Koivisto et al., 2007; Qiao et al., 2008). This therapies for PrCa. These may be particularly attractive may be because of critical transcriptional co-activator in subsets of tumors that show overexpression of the functions of PHF8 in the regulation of neuronal and respective protein targets and pathways, such as PHF8 craniofacial development (Fortschegger et al., 2010; nuclear overexpression. Larger scale efforts of mapping Kleine-Kohlbrecher et al., 2010; Qi et al., 2010). epigenomic landscapes in cancer (Abbott, 2010), com- Furthermore, we showed that PHF8 mRNA is over- bined with the functional understanding of the function expressed in a large fraction of PrCa and correlated of epigenetic enzymes, may help to identify the most with features of tumor progression such as high Gleason feasible and cancer-specific epigenetic therapy formats grading. This correlation with adverse events and in the future. poor patient survival was further validated by immunohistochemistry staining of TMAs from clinical samples. Immunohistochemical staining of PrCa tissues Materials and methods showed that PHF8 localization was preferentially nucleolar, compatible with the assumed function of an Cell lines epigenetic modifier and the previously reported role in The prostate cell lines RWPE-1, MDA PCa 2b, DU-145 LNCaP regulation of ribosomal RNA promoters (Feng et al., and PC-3 were purchased from ATCC (LGC Promochem AB, 2010). Enhanced activity of ribosomal gene expression is Boras, Sweden), and grown in the media recommended by the a typical observation in hyper-proliferative cancers. distributor. The prostate carcinoma cell line VCaP was received Both the analysis of putative PHF8-regulated genes in from Adrie van Bokhoven (University Medical Center Nijmegen, Netherlands). All tumor cell lines were grown in the RPMI-1640 the LNCaP cell line, and the genes correlating with medium. All media were supplemented with 1% (for RWPE1, PHF8 expression in primary PrCa showed an enrich- prostate epithelial cells) or 10% FCS (all cancer lines), ment of GO categories related to the actin cytoskeleton L-glutamine and penicillin/streptomycin (all reagents from and integrin signaling. Time-lapse microscopy and 3-D Sigma, Munich, Germany). All cells were cultured for o4 organotypic cell cultures showed that PHF8 is also a months before use in these experiments.

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3453 Cell-spot microarrays manufacturer’s instruction. The results were compared with Cell microarray screens were performed, normalized and mock transfected cells (empty vector). analyzed as previously described in (Rantala et al., 2010 and Rantala et al., 2011). Briefly, the epigenomic siRNA library (Qiagen, Hilden, Germany) was complexed with lipid transfec- Statistical analysis of the histone demethylase siRNA screening tion agents and Matrigel (BD Biosciences, San Jose, CA, USA) The secondary screening focused on histone demethylases and and spotted/arrayed onto an untreated hydrophobic poly- JmjC-domain containing proteins, identified as significantly styrene surface. VCaP cells were applied onto the array as a enriched in the primary screen. Four siRNAs per gene were suspension, and allowed to adhere for 15 min. Unattached cells used. To normalize the siRNA screening data the loess were then washed off. After 48 h cells were fixed with 2% function from the R-package ‘stats’, using row and column paraformaldehyde and permeabilized with 0.3% Triton-X 100 coordinates as predictors, was fitted across each 384-well plate. and 10% horse serum in PBS, and blocked with 10% horse In order to achieve a robust fit, statistical outliers were serum. Arrays were assayed with antibodies for Ki-67 (Abcam, estimated before the fitting was carried out and their weight Cambridge, UK), cleaved PARP (Cell Signalling Technolo- was set to 10% of the normal weight. The fit was then gies, Danvers, MA, USA), AR (Thermo Scientific, Waltham, subtracted from the original values, the data were divided by MA, USA), histone H3 lysine 4 dimethylation (Abcam) and the median of negative controls on the plate, and log2 H3K9me2 (Abcam), and labeled with goat-anti-mouse and transformed. All siRNAs reducing cell viability by at least donkey-anti rabbit Alexa488, À555 or À647 dyes (Molecular 1.8 s.d. from the median of the controls (corresponding to Probes, Carlsbad, CA, USA). Immunostained cell microarrays Po0.05) were considered as anti-proliferative hits (Supple- were analyzed by microscopic imaging, using an Olympus mentary Table 4). Each screen was analyzed using the scanR high content imager (Olympus, Hamburg, Germany) redundant siRNA analysis algorithm and P-values were equipped with a Hamamatsu ORCA-ER CCD digital camera assigned to each gene. The redundant siRNA analysis P-value (Hamamatsu Photonics K.K., Hamamatsu, Japan). Spot represents the likelihood of the corresponding siRNA signal images were analyzed using the scanR image analysis software distribution to be generated by chance. Hits confirmed by the (scanR Inc., Palo Alto, CA, USA). Total raw mean intensities redundant siRNA analysis method at level of Po0.05 were of stains for all cells per spot were normalized using pin used in validation experiments. normalization. For analysis of significance, a Z-score was calculated for all measured parameters of each spot using Immunoblotting global array mean value and s.d. for all samples in the array, Aliquots of total cell lysates were fractionated on SDS including controls. From this analysis, siRNAs with polyacrylamide gels and transferred to Whatman Protran Z-scores±2 s.d. were considered significant. To assay all nitrocellulose membranes (Schleicher & Schuell, Whatman seven biological endpoints, four independent replicate arrays Inc., Florham Park, NJ, USA). Membranes were probed with with dual staining of the markers were used. a specific antibody for PHF8 (Abcam ab35471) and equal loading was confirmed with human b-actin (Sigma-Aldrich). siRNA protein class-based pattern identification analysis Signal was detected with Alexa Fluor-680 goat anti-rabbit and For analysis of significance, a Z-score was calculated for all Alexa Fluor-680 conjugated rabbit anti-goat secondary anti- measured parameters of each array spot using global array bodies (Invitrogen Molecular Probes), and scanned with an median value and s.d. for all samples in the array, including Odyssey infrared scanner (Li-Cor Biosciences, Lincoln, NE, the controls. From the analysis, siRNAs with Z-scores±2 s.d. USA). greater or less than the global array median were considered significant. The partitioning around medoids method as Quantitative RT–PCR implemented in the R package cluster was used for dividing Total cellular RNA was isolated with the RNeasy kit (Qiagen). the siRNAs exceeding the threshold in any of the seven For cDNA synthesis, 200 ng of total RNA was reverse markers into six clusters. The R statistical programming transcribed with a High Capacity cDNA Reverse Transcrip- language was employed for hierarchical clustering within the tion kit (Applied Biosystems, Foster City, CA, USA). There- clusters. The DAVID online GO-mining tool and the EASE after, the cDNAs were diluted 1/10 and Taqman quantitative score (Hosack et al., 2003) were employed to identify Interpro real-time–PCR analysis was carried out with an Applied domains and protein families enriched in clusters with cancer- Biosystems 7900HT instrument using specific primers for relevant properties (Supplementary Table 2). The siRNAs PHF8 and b-actin designed by the Universal Probe Library included in the secondary screening were selected based on the Assay Design Center (Roche, Basel, Switzerland). The enrichment analysis results. sequences were as follows: PHF8 forward 50-GCCTCTATTG AGACAGGTTTGG-30, PHF8 reverse 50-TCTTTTGGGCCT TCTGTAGC-30, b-actin forward 50-CCAACCGCGAGAAG siRNA and DNA transfections ATGA-30 and b-actin reverse 50-CCAGAGGCGTACAGGG SiRNA molecules targeting the gene of interest were trans- ATAG-30. The fluorescent Taqman probes were obtained from fected by using reverse transfection with siLentFect transfec- Roche Human Probe Library (no 3 for PHF8 and no 64 for b- tion reagent (Bio-Rad Laboratories, Hercules, CA, USA) in actin). The results were analyzed with SDS 2.3 and RQ Opti-MEM (Invitrogen, Carlsbad, CA, USA). All target genes manager software (Applied Biosystems), and the expression of and siRNAs used are listed in Supplementary Tables 1 and 3. PHF8 mRNA was determined by the relative quantitation AllStars Negative Control (Qiagen) was used as scrambled method using b-actin as an endogenous control. Data were control. The final siRNA concentration used was 13 nmol/l. collected from two separate biological experiments, which were HA-tagged PHF8 expression vector was a generous gift from both run twice with triplicates. Professor Kristian Helin (Biotech Research and Innovation Centre, University of Copenhagen, Denmark). 1 mg of DNA was transfected into RWPE1 and PC-3 cells grown at 70% Gene expression analysis with Illumina bead-arrays confluency, using the FuGeneHD transfection reagent Early passage LNCaP cells transfected with scramble and (Applied Biosciences, Basel, Switzerland), according to the PHF8 siRNAs for 48 h and total RNA was extracted using

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3454 RNeasy Mini kit (Qiagen). Integrity of the RNA before Tissue microarray construction and immunohistochemistry hybridization was monitored using a Bioanalyzer 2100 TMA2 blocks were constructed using archival formalin-fixed (Agilent, Santa Clara, CA, USA). Purified total RNA paraffin-embedded prostatectomy blocks. Recipient blocks were (500 ng) was amplified with the TotalPrep Kit (Ambion, pre-drilled with an automated TMA instrument (TMA Master, Austin, TX, USA) and the biotin labeled cRNA was 3D Histech, Budapest, Hungary) and donor block cores (+ hybridized to Sentrix HumanRef-8 Expression BeadChips 1.0 mm) were manually punched (Tissue-Tek Quick-RayTM, (Illumina, San Diego, CA, USA). The arrays were scanned Sakura Finetekk Europe B.V., Alphen aan den Rijn, The with the BeadArray Reader (Illumina). The raw data were Netherlands). To account for tumor heterogeneity, at least two quantile-normalized and analyzed with the R/Bioconductor cores were drilled from the area containing the most dominant software and beadarray package (Dunning et al., 2007). Gleason grade pattern and one core from the area containing the Differentially expressed genes from two independent micro- second most dominant Gleason pattern. In addition, one core of array hybridizations were defined to be X1.4-fold under- or each patient contained an adjacent benign glandular area. All overexpressed with a P-value o0.01 (Supplementary Table 7). cancer cores in the TMA were graded individually, according to The functional GO and pathway annotations were analyzed contemporary Gleason grading criteria. The TMA blocks for sets of differentially expressed genes by EASE score (http:// contained a total of 1478 cores. Freshly cut TMA sections were niaid.abcc.ncifcrf.gov) (Hosack et al., 2003). mounted on electrically charged glass slides (SuperFrost Plus, Menzel Glaeser, Braunschweig, Germany) and stained with a fully automated immunohistochemical system Benchmark XT Cell proliferation assays (Ventana Medical Systems, Illkirch, France), using a biotin-free Cell proliferation assays were performed on 384-well plates multimer-based detection system (ultraView Universal Red, with optimized cell density, plated in their respective growth Ventana Medical Systems). The slides were pre-treated with media. SiRNA and lipid mixture were incubated on the plates 1 h standard cell conditioning, using CC1 buffer and incubated for at RT, then cells were added and incubated for 72 h. Cell viability 60 min with the rabbit polyclonal antibody to PHF8 (1:250, was measured by CellTiter-Glow (Promega, Madison, WI, Abcam). Hematoxylin II and Blueing Reagent (Ventana Medical USA), according to the manufacturer’s instructions. The Systems) were used as counterstains. Negative controls were Envision Multilabel Plate Reader (Perkin-Elmer, MA, USA) performed by replacing the primary antibody with corresponding was used for luminescent signal quantification. The mean values rabbit IgG. The TMA slides were digitized with an automated of at least three independent measurements were used and whole slide scanner (Mirax Scan, Zeiss, Go¨ttingen, Germany), Student’s t-test applied for calculating the significance of changes. usinga20Â objective (numerical aperture 0.75) and a Sony DFW-X710 camera with a 1024 Â 768 pixel CCD sensor (Sony Cell cycle analysis Corporation, Tokyo, Japan). The images were compressed to a Cell cycle analysis was carried out for scramble and siPHF8 wavelet file format (Enhanced Compressed Wavelet (ECW), ER transfected LNCaP cells. Cells were stained with propidium Mapper, Erdas Inc, Atlanta, GA, USA) with a conservative iodine staining after 72 h incubations followed by flow cytometry compression ratio of 1:5. The compressed virtual slides were analysis using the BD FACSArray Bioanalyzer (Becton & uploaded to a web server (http://fimm.webmicroscope.net) Dickinson, Franklin Lakes, NJ, USA). Samples were collected, running image server software (Image Web Server, ERDAS stained and analyzed according to the manufacturer’s instruc- Intergraph Inc., Madison, AL, USA). Reliable Gleason grading tions. Growth curves in the IncuCute real-time imaging device was possible for 1179 of the 1478 TMA cores stained with PHF8 were generated as described (Harma et al., 2010). antibody, and patient overall survival data was available for 332 patients. In the survival analysis, a maximum score of each cancer cores of an individual patient was correlated with the mRNA expression in clinical samples clinical end point. Expression of PHF8 was evaluated by two Analysis of relative expression level of candidate genes in the pathologists (TM/AS) independently from the digitized slides GeneSapiens database was performed as described previously described above. The investigators were blinded to the clinico- (Kilpinen et al., 2008). Briefly; relative expression level of pathological data at the time of scoring. Nuclear/nucleolar candidate genes across 496 clinical human prostate tissue staining intensity was scored as follows: no staining (negative), samples, consisting of 147 normal samples and 349 tumors and weak ( þ ), moderate ( þþ)orstrong(þþþ) staining. metastases was analyzed and the statistical significance of the difference in gene expression between these sample groups and Cell migration correlation of PHF8 expression with other genes was confirmed Cell migration was measured with time-lapse microscopy as t with -statistics (Supplementary Tables 5 and 6). Other data sets previously described (Tuomi et al., 2009). Briefly, PC-3 cells were downloaded from gene expression omnibus or the MSKCC were transfected for 48 h with either mock, HA-PHF8, server, normalized and mined using the same online tools scramble or PHF8 siRNA, and cell migration was imaged et al. (R bioconductor) as described (Kilpinen , 2008). from eight random fields per treatment for 16 h. Phase contrast images were taken with a Zeiss inverted wide-field microscope Clinical PrCa samples (EL Plan-Neofluar 20 Â /0.5 NA objective, 12 frames per In all, 378 PrCa patients treated with total prostatectomy hour), equipped with a heated chamber (37 1C) and CO2 between the years 1982 and 1998 at the Helsinki University controller (4.8%). Image processing was performed with Central Hospital, Finland, were included in this study. Median MetaMorph Offline software (Molecular Devices, Sunnyvale, age of the patients at diagnosis was 63 (44–83) years. None of CA, USA). Length of the migration of 15 to 20 cells per the patients had received adjuvant therapy before or immedi- treatment was quantified in pixels. ately after the surgery. Histopathological features were reviewed using the corresponding haematoxylin and eosin or 3-D invasion assays herovici-stained slides by two experienced pathologists, and PC-3 cells were reverse-transfected for 68 h using SiLentFect clinical followup information was collected from patient files. lipid transfection reagent (Bio-Rad Laboratories, Hercules, All tissue samples were acquired and used according to CA, USA). Transfected cells were embedded between two contemporary regulatory guidelines. layers of Matrigel on Angiogenesis slides (ibidi GmbH) (1500

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3455 cellsper well) as described before in Harma et al. (2010) and Conflict of interest cultured for 8 days. On day 8, the multicellular structures were visualized with SYTO62 DNA dye (Invitrogen) and NucView The authors declare no conflict of interest. 488 caspase detection reagent (Biotium Inc., Hayward, CA, USA). 3-D image stacks were acquired with a Zeiss spinning disk confocal microscope as described in Harma et al. (2010). Maximum intensity projections were generated and normal- Acknowledgements ized with Slidebook 5.0 software (Intelligent Imaging Innova- tions (3i) Inc., Denver, CO, USA). Remaining background We thank Professor Tony Kouzarides from The Gurdon noise was removed with ImageJ (National Institute of Health, Institute, University of Cambridge for valuable suggestions Bethesda, MD, USA). Image projections were then analyzed and insights in constructing the epigenetic siRNA library. We with VTT’s proprietary automated image analysis program thank Daniela Kleine-Kohlbrecher and Professor Kristian software (v1.95). Automated image analysis program segments Helin from Biotech Research and Innovation Centre (BRIC), individual cell structures based on morphological criteria, and University of Copenhagen, Denmark, for insightful discus- assigns numerical values for selected morphological features sions, sharing unpublished results and PHF8 over expression including area (size), roundness (shape) and roughness construct. We thank Arho Virkki and Elmar Bucher for (contour). VTT automated image analysis program is specifi- assistance with gene expression data normalization. This study cally designed to measure the number and length of invasive was supported by grants from Academy of Finland (grant Nr. cellular structures (median, index and maximum extension of 111456 to MN, VH), EU-EPITRON (contract # LSHC-CT- ‘appendages’). Apoptosis is indicated as ratio of NucView 2005-518417 to MB, OK), Marie Curie Canceromics (MEXT- signal to structure size. Raw numerical data were further CT-2003-2728 to OK), Cancer Organizations Finland and statistically processed and visualized with R/Bioconductor. Sigrid Juselius Foundation (JPM, JR).

References

Abbott A. (2010). Project set to map marks on genome. Nature 463: Feng W, Yonezawa M, Ye J, Jenuwein T, Grummt I. (2010). PHF8 596–597. activates transcription of rRNA genes through H3K4me3 binding Abidi FE, Miano MG, Murray JC, Schwartz CE. (2007). A novel and H3K9me1/2 demethylation. Nat Struct Mol Biol 17: 445–450. mutation in the PHF8 gene is associated with X-linked mental Flamand V, Zhao H, Peehl DM. (2010). Targeting monoamine oxidase A retardation with cleft lip/cleft palate. Clin Genet 72: 19–22. in advanced prostate cancer. J Cancer Res Clin Oncol 11: 1761–1771. Baker LA, Allis CD, Wang GG. (2008). PHD fingers in human Fortschegger K, de Graaf P, Outchkourov NS, van Schaik FM, diseases: disorders arising from misinterpreting epigenetic marks. Timmers HT, Shiekhattar R. (2010). PHF8 targets histone Mutat Res 647: 3–12. methylation and RNA polymerase II to activate transcription. Bianco-Miotto T, Chiam K, Buchanan G, Jindal S, Day TK, Thomas Mol Cell Biol 30: 3286–3298. M et al. (2010). Global levels of specific histone modifications and Fraga MF, Ballestar E, Villar-Garea A, Boix-Chornet M, Espada J, an epigenetic gene signature predict prostate cancer progression and Schotta G et al. (2005). Loss of acetylation at Lys16 and development. Cancer Epidemiol Biomarkers Prev 19: 2611–2622. trimethylation at Lys20 of histone H4 is a common hallmark of Birmingham A, Selfors LM, Forster T, Wrobel D, Kennedy CJ, human cancer. Nat Genet 37: 391–400. Shanks E et al. (2009). Statistical methods for analysis of high- Guo W, Giancotti FG. (2004). Integrin signalling during tumour throughput RNA interference screens. Nat Methods 6: 569–575. progression. Nat Rev Mol Cell Biol 5: 816–826. Bjorkman M, Iljin K, Halonen P, Sara H, Kaivanto E, Nees M et al. Halkidou K, Gaughan L, Cook S, Leung HY, Neal DE, Robson CN. (2008). Defining the molecular action of HDAC inhibitors and (2004). Upregulation and nuclear recruitment of HDAC1 in synergism with androgen deprivation in ERG-positive prostate hormone refractory prostate cancer. Prostate 59: 177–189. cancer. Int J Cancer 123: 2774–2781. Harma V, Virtanen J, Makela R, Happonen A, Mpindi JP, Knuuttila Chandran UR, Ma C, Dhir R, Bisceglia M, Lyons-Weiler M, Liang W M et al. (2010). A comprehensive panel of three-dimensional models et al. (2007). Gene expression profiles of prostate cancer reveal for studies of prostate cancer growth, invasion and drug responses. involvement of multiple molecular pathways in the metastatic PLoS One 5: e10431. process. BMC Cancer 12: 64. Holst CR, Nuovo GJ, Esteller M, Chew K, Baylin SB, Herman JG Cloos PA, Christensen J, Agger K, Maiolica A, Rappsilber J, Antal T et al. (2003). Methylation of p16(INK4a) promoters occurs in vivo in et al. (2006). The putative oncogene GASC1 demethylates tri- and histologically normal human mammary epithelia. Cancer Res 63: dimethylated lysine 9 on histone H3. Nature 442: 307–311. 1596–1601. Cooper CS, Foster CS. (2009). Concepts of epigenetics in prostate Horton JR, Upadhyay AK, Qi HH, Zhang X, Shi Y, Cheng X. (2010). cancer development. Br J Cancer 100: 240–245. Enzymatic and structural insights for substrate specificity of a Dunning MJ, Smith ML, Ritchie ME, Tavare S. (2007). Beadarray: R family of jumonji histone lysine demethylases. Nat Struct Mol Biol classes and methods for illumina bead-based data. Bioinformatics 17: 38–43. 23: 2183–2184. Hosack DA, Dennis G, Sherman BT, Lane HC, Lempicki RA. (2003). Egger G, Liang G, Aparicio A, Jones PA. (2004). Epigenetics in human Identifying biological themes within lists of genes with EASE. disease and prospects for epigenetic therapy. Nature 429: 457–463. Genome Biol 4: R70. Ellinger J, Kahl P, von der Gathen J, Rogenhofer S, Heukamp LC, Kahl P, Gullotti L, Heukamp LC, Wolf S, Friedrichs N, Vorreuther R Gutgemann I et al. (2010). Global levels of histone modifications et al. (2006). Androgen receptor coactivators lysine-specific histone predict prostate cancer recurrence. Prostate 70: 61–69. demethylase 1 and four and a half LIM domain protein 2 predict Ellis L, Atadja PW, Johnstone RW. (2009). Epigenetics in cancer: risk of prostate cancer recurrence. Cancer Res 66: 11341–11347. targeting chromatin modifications. Mol Cancer Ther 8: 1409–1420. Kilpinen S, Autio R, Ojala K, Iljin K, Bucher E, Sara H et al. (2008). Ernst J, Kellis M. (2010). Discovery and characterization of chromatin Systematic bioinformatic analysis of expression levels of 17,330 states for systematic annotation of the human genome. Nat human genes across 9,783 samples from 175 types of healthy and Biotechnol 28: 817–825. pathological tissues. Genome Biol 9: R139.

Oncogene PHF8 mediates cell migration in prostate cancer MBjo¨rkman et al 3456 Kleine-Kohlbrecher D, Christensen J, Vandamme J, Abarrategui I, Seligson DB, Horvath S, McBrian MA, Mah V, Yu H, Tze S et al. Bak M, Tommerup N et al. (2010). A functional link between the (2009). Global levels of histone modifications predict prognosis in histone demethylase PHF8 and the transcription factor ZNF711 in different cancers. Am J Pathol 174: 1619–1628. X-linked mental retardation. Mol Cell 10: 1016. Seligson DB, Horvath S, Shi T, Yu H, Tze S, Grunstein M et al. Koivisto AM, Ala-Mello S, Lemmela S, Komu HA, Rautio J, Jarvela (2005). Global histone modification patterns predict risk of prostate I. (2007). Screening of mutations in the PHF8 gene and identifica- cancer recurrence. Nature 435: 1262–1266. tion of a novel mutation in a finnish family with XLMR and cleft Sharma S, Kelly TK, Jones PA. (2010). Epigenetics in cancer. lip/cleft palate. Clin Genet 72: 145–149. Carcinogenesis 31: 27–36. Lange SS, Mitchell DL, Vasquez KM. (2008). High mobility group Shi X, Kachirskaia I, Walter KL, Kuo JH, Lake A, Davrazou F et al. protein B1 enhances DNA repair and chromatin modification after (2007). Proteome-wide analysis in saccharomyces cerevisiae identi- DNA damage. Proc Natl Acad Sci USA 105: 10320–10325. fies several PHD fingers as novel direct and selective binding Laumonnier F, Holbert S, Ronce N, Faravelli F, Lenzner S, modules of histone H3 methylated at either lysine 4 or lysine 36. Schwartz CE et al. (2005). Mutations in PHF8 are associated with J Biol Chem 282: 2450–2455. X linked mental retardation and cleft lip/cleft palate. JMedGenet42: Shin S, Janknecht R. (2007). Activation of androgen receptor by 780–786. histone demethylases JMJD2A and JMJD2D. Biochem Biophys Res Lee MG, Wynder C, Schmidt DM, McCafferty DG, Shiekhattar R. Commun 359: 742–746. (2006). Histone H3 lysine 4 demethylation is a target of nonselective Smith EH, Janknecht R, Maher III LJ. (2007). Succinate inhibition of antidepressive medications. Chem Biol 13: 563–567. alpha-ketoglutarate-dependent enzymes in a yeast model of para- Liu W, Tanasa B, Tyurina OV, Zhou TY, Gassmann R, Liu WT et al. ganglioma. Hum Mol Genet 16: 3136–3148. (2010). PHF8 mediates histone H4 lysine 20 demethylation events Tang D, Kang R, Zeh III HJ, Lotze MT. (2010). High-mobility group involved in cell cycle progression. Nature 466: 508–512. box 1 and cancer. Biochim Biophys Acta 1799: 131–140. Loenarz C, Ge W, Coleman ML, Rose NR, Cooper CD, Klose RJ Taylor BS, Schultz N, Hieronymus H, Gopalan A, Xiao Y, Carver BS et al. (2010). PHF8, a gene associated with cleft lip/palate and et al. (2010). Integrative genomic profiling of human prostate mental retardation, encodes for an nepsilon-dimethyl lysine cancer. Cancer Cell 18: 11–22. demethylase. Hum Mol Genet 19: 217–222. Tuomi S, Mai A, Nevo J, Laine JO, Vilkki V, Ohman TJ et al. Mellor J. (2006). It takes a PHD to read the histone code. Cell 126: (2009). PKCepsilon regulation of an alpha5 integrin-ZO-1 complex 22–24. controls lamellae formation in migrating cancer cells. Sci Signal Metzger E, Wissmann M, Yin N, Muller JM, Schneider R, Peters AH 2: ra32. et al. (2005). LSD1 demethylates repressive histone marks to promote Weichert W, Roske A, Gekeler V, Beckers T, Stephan C, Jung K et al. androgen-receptor-dependent transcription. Nature 437: 436–439. (2008). Histone deacetylases 1, 2 and 3 are highly expressed in Minucci S, Pelicci PG. (2006). Histone deacetylase inhibitors and the prostate cancer and HDAC2 expression is associated with promise of epigenetic (and more) treatments for cancer. Nat Rev shorter PSA relapse time after radical prostatectomy. Br J Cancer Cancer 6: 38–51. 98: 604–610. Qi HH, Sarkissian M, Hu GQ, Wang Z, Bhattacharjee A, Welsbie DS, Xu J, Chen Y, Borsu L, Scher HI, Rosen N et al. (2009). /delim>Gordon DB et al. (2010). Histone H4K20/H3K9 demethy- Histone deacetylases are required for androgen receptor function in lase PHF8 regulates zebrafish brain and craniofacial development. hormone-sensitive and castrate-resistant prostate cancer. Cancer Nature 466: 503–507. Res 69: 958–966. Qiao Y, Liu X, Harvard C, Hildebrand MJ, Rajcan-Separovic E, Wissmann M, Yin N, Muller JM, Greschik H, Fodor BD, Jenuwein T Holden JJ et al. (2008). Autism-associated familial microdeletion of et al. (2007). Cooperative demethylation by JMJD2C and LSD1 Xp11.22. Clin Genet 74: 134–144. promotes androgen receptor-dependent gene expression. Nat Cell Rantala JK, Edgren H, Lehtinen L, Wolf M, Kleivi K, Aaltola AR Biol 9: 347–353. et al. (2010). Integrative functional genomics analysis of sustained Wolf SS, Patchev VK, Obendorf M. (2007). A novel variant of the polyploidy phenotypes in breast cancer cells identifies an oncogenic putative demethylase gene, s-JMJD1C, is a coactivator of the AR. profile forGINS2. Neoplasia 12: 877–888. Arch Biochem Biophys 460: 56–66. Rantala JK, Makela R, Aaltola AR, Laasola P, Mpindi JP, Nees M Xiang Y, Zhu Z, Han G, Ye X, Xu B, Peng Z et al. (2007). JARID1B et al. (2011). A cell spot microarray method for production of is a histone H3 lysine 4 demethylase up-regulated in prostate cancer. high density siRNA transfections microarrays. BMC Genomics 12: Proc Natl Acad Sci USA 104: 19226–19231. 162–176. Yamane K, Toumazou C, Tsukada Y, Erdjument-Bromage H, Tempst Rose NR, Ng SS, Mecinovic J, Lienard BM, Bello SH, Sun Z et al. P, Wong J et al. (2006). JHDM2A, a JmjC-containing H3K9 (2008). Inhibitor scaffolds for 2-oxoglutarate-dependent histone demethylase, facilitates transcription activation by androgen lysine demethylases. J Med Chem 51: 7053–7056. receptor. Cell 125: 483–495. Ruthenburg AJ, Allis CD, Wysocka J. (2007). Methylation of lysine 4 Zhu Z, Wang Y, Li X, Wang Y, Xu L, Wang X et al. (2010). PHF8 is a on histone H3: Intricacy of writing and reading a single epigenetic histone H3K9me2 demethylase regulating rRNA synthesis. Cell Res mark. Mol Cell 25: 15–30. 20: 794–801.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene